Skip to content
AI Engineering Knowledge BaseAI Engineering Knowledge Base

Anti-Patterns

Common Anti-Patterns

Section titled “Common Anti-Patterns”
  • Treating an AI model as the whole product instead of one system component.
  • Measuring aggregate model quality without qualitative error analysis.
  • Letting agent-generated code bypass architecture, tests or human review.
  • Adding MCP tools or write permissions without clear contracts and trust boundaries.
  • Starting from an impressive AI feature rather than a shaped user problem.
  • Using human approval as a symbolic control when reviewers lack time, authority or information.